TR2009-034

Multi-Class Active Learning for Image Classification


Abstract:

One of the principal bottlenecks in applying learning techniques to classification problems is the large amount of labeled training data required. Especially for image and video, providing training data is very expensive in terms of human time and effort. In this paper we propose an active learning approach to tackle the problem. Instead of passively accepting random training examples, the active learning algorithm iteratively selects unlabeled examples for the user to label, so that human effort is focused on labeling the most \"useful\" examples. Our method relies on the idea of uncertainty sampling, in which the algorithm selects unlabeled examples that it finds hardest to classify. Specifically, we propose an uncertainty measure that generalizes margin-based uncertainty to the multi-class case and is easy to compute, so that active learning can handle a large number of classes and large data sizes efficiently. We demonstrate results for letter and digit recognition on datasets from the UCI repository, object recognition results on the Caltech-101 dataset, and scene categorization results on a dataset of 13 natural scene categories. The proposed method gives large reductions in the number of training examples required over random selection to achieve similar classification accuracy, with little computational overhead.

 

  • Related News & Events

    •  NEWS    CVPR 2009: 6 publications by Amit Agrawal and others
      Date: June 20, 2009
      Where: IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
      Research Area: Computer Vision
      Brief
      • The papers "3D Pose Estimation and Segmentation using Specular Cues" by Chang, J.-Y., Raskar, R. and Agrawal, A., "Coded Exposure Deblurring: Optimized Codes for PSF Estimation and Invertibility" by Agrawal, A. and Xu, Y., "Enforcing Integrability by Error Correction using $l_1$-minimization" by Reddy, D., Agrawal, A. and Chellappa, R., "Multi-Class Active Learning for Image Classification" by Joshi, A.J., Porikli, F. and Papanikolopoulos, N., "Optimal Single Image Capture for Motion Deblurring" by Agrawal, A. and Raskar, R. and "Geometric Sequence (GS) Imaging with Bayesian Smoothing for Optical and Capacitive Imaging Sensors" by Sengupta, K. and Porikli, F. were presented at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR).
    •